Latest / The MonkCast / An Interview with Kyle Daigle at Microsoft BUILD 2026: Models, Tokens, Agents & Developer Calm

Transcript

An Interview with Kyle Daigle at Microsoft BUILD 2026: Models, Tokens, Agents & Developer Calm · The MonkCast

• speaker-0: Hey, it's James Governor, co-founder of Redmunk. ⁓ I'm here today ⁓ at Build, Microsoft Build 2026 in Fort Mason in San Francisco. I'm here with Kyle Daigle, the ⁓ developer chief marketing officer for Microsoft. And ⁓ yeah, we're here to talk a little bit about the event, ⁓ perhaps my reckons, what you think people should be taking away from the event. ⁓ so just to kick off, what is a developer chief marketing officer, Kyle? speaker-1: Yeah. So I mean, you know, I've ⁓ I've been a leader at GitHub for quite some time as, you know, and I think for me the goal in this role is how can we take what developers care about? I mean, like you have to be able to use a tool, you have to be able to understand it, you have to be able to ⁓ like we have to be authentic about what these things can or can't do. And it's my job to bring our teams together across all the products at Microsoft. To ensure that we're coming to you not with, you know, four or five products on a shelf as much as we can. And instead saying, hey, if you're a developer and you're trying to build software, you know, using agents, here's the complete flow that you need with the ability to go use those products with the documentation, kind of so on and so forth. So developers can't be marketed to in the same way, you know, that other, I think, you know, buyers can. but it ⁓ it is about just, you know. being developer first and making sure that, you know, my teams as well as the engineering teams, the product teams, the sales teams know what that means and how we can partner on that together. speaker-0: So ⁓ I to be fair, ⁓ I'll I'll ⁓ I I will say that having arrived ⁓ at ⁓ at this you know venue Fort Mason, I'm I've been here for GitHub universe before. and I did very in very quickly get the impression I thought I thought that build feels, you know, very I I've got here this morning, I'm like, Build feels a little bit more developer y than usual. So you're hitting the mark. Well done. ⁓ there was a lot, a lot speaker-1: Yeah. speaker-0: of ⁓ news today. Yeah. ⁓ mid two and a half hour keynote. Flew by. Flew by. Because there was so much news. ⁓ but certainly plenty to ⁓ contextualize. ⁓ I've got some things that I think probably struck me. Yeah. ⁓ so in fact we'll jump in with that. ⁓ speaker-1: ⁓ speaker-0: Microsoft has its own frontier models now. So ⁓ assuming the benchmarks are reflected in reality, it looks like with these new My Models, you're really putting a marker down in terms of the ability to ⁓ reason, do text, and code, most importantly in terms of what we're about. Tell me a bit about that. When you're trying to get the world to understand that why is it so important that Microsoft has its own model? speaker-1: Yeah. I mean, I think that, you know, Mustafa said it best today. It's I think they're taking an approach where they're starting or they started, you know, kind of from zero ⁓ and you know, didn't use distillation. They're not building on top of another model. They're starting from scratch. And there's an advantage of that, I think, for ⁓ the ability to use these models in unique ways at work, you know, being able to ⁓ you you know fine-tune these models and know exactly sort of what you're starting with. I think the thing that is more exciting for me is the fact that It's starting at a place that's like grounded in our current token and operational reality. We don't need absolutely everyone to be going after these models that are enormous frontier models that cost a lot per model, going into it in a place of saying, hey, performance and what these things can do and how much they're gonna cost starting there. Token efficiency ⁓ is exciting to me because we're already seeing the kind of push and pull of not everything. that can be sent down these more expensive models at some point I get the bill at the end of the month. I mean like it's you know even for myself, you know, like if I'm using my ⁓ personal open quads and calling these ⁓ APIs, it's like I I don't know that I needed that much help to find the restaurant phone number, you know. and I think the MAI models, the thinking model in particular, the Code One model, ⁓ feels like a speaker-0: You're done. speaker-1: really, really great job of that. The transcribe models, what's powering the, you know, the co-pilot CLI voice. ⁓ and so it's functionality, but with like some rationalization of the world that we're in and that ⁓ performance and cost are s also a factor, not just ⁓ what it can do. speaker-0: Okay. I mean I I I f off the build last year, I said the Microsoft needs its own frontier models. It needs that moonshot. ⁓ you you you can't be in a position where you're beholden ⁓ to open AI and to anthropic. Obviously they are crushing it in market. ⁓ I mean your positioning there is a bit more about the economics, which is super important, but obviously you've got to have ambitions ⁓ that the model is the models will be competitive ⁓ with what we're getting from those providers. speaker-1: Yeah, I mean as like a first ⁓ as a first showing, you know, for Mustafa to be able to have code one and get it to a four-six sonnet as like the first attempt, I think, is really strong. And you can see the hill climbing chart that he shared, and they're actually like accelerating the speed at which they're able to bring intelligence to these models. So this is the starting line to some degree, you know, of this new era of models. But to be clear, right? The ambition is not just have, you know, ⁓ we're the most economic model or like, you know, we're the ones that are taking in performance solely. I think it's just about being clear that we can have this really frontier intelligence and be cognizant of how do we do that with token efficiency. You know, and not just look at it from a single ⁓ a single place. speaker-0: the next the next era frankly is going to be defined by I mean I've been calling it token economics talking about token efficiency. ⁓ we can't keep on burning money in the way that we have been ⁓ that's very clear. So ⁓ you know I I think that that that focus ⁓ in the in the keynote today it was it that that that was I mean there are two aspects of this Microsoft perspective. One is that full stack ⁓ running on your own chips. You've got your own model now, you're not beholden to third parties, first party models. ⁓ the other one, of course, is this this question about running ⁓ running things locally. Yeah. ⁓ which should be a Microsoft advantage. And today I I guess that was one of the the key themes. What tell me about that and whether that's realistic. I mean, if i the the the the perhaps the inner loop and the ability to use a model locally. speaker-1: Yeah, I mean, ⁓ I think there's kind of the the chip aspect of it first, you know. ⁓ the N one X Spark and ⁓ I've been able to use the laptop myself and actually being able to use the models that have been coming out ⁓ locally is such a huge boon. I think it's a boon not just for, you know, ⁓ when I'm working on problem sets that are, you know Frontier problems, you know, being able to just have that sort of speed or have it in the background has been quite helpful to me. There's been a variety of open source models. If we've been using like a ⁓ quen in some places for this, ⁓ there's like internal models that we're testing out. And of course, like the the the SLM side does this, but developers are looking for can I run an actual Mars large-ish language model, 120 billion parameter model ⁓ on my laptop? And we're I think I'm most excited is not just having that on device, but also having the Microsoft execution containers on device because I would love to be able to run a large language model on my laptop and be able to run a claw that can be used at work because it's fully contained and observable at a work setting. And that is, I think, the unique opportunity that Microsoft is like aiming at is that I wanna have control. Like I wanna be able to work like I do on Saturday, but at work. ⁓ and threading that needle where like the devs get to hear about, you know, bringing claw to Windows and the Windows Companion app, et cetera, like that's great. And then, you know, enterprises can hear, ⁓ well, we have containment, we have observability for your claws or local agents. And then everyone can run it locally. The models are getting so good, whether you're using a quantized one or not, like they're very good locally with this device in particular. speaker-0: The claw stuff, I've got to admit, I mean, I'm I'm I'm I you know I guess I'm I'm kind of a boomer. ⁓ I get I get I get very nervous ⁓ about ⁓ well just the the the capability that's there, right? ⁓ I thought it was very interesting the degree to which you lent into that on on i i in the in in in the the keynote where ⁓ you know ⁓ the conceit is, you know, we're gonna get the claw to delete everything on my desktop. You know, everyone's like, No. But but ⁓ yeah, having containment. Yeah. ⁓ so that doesn't happen. Yep. ⁓ so you had ⁓ Steinberger here, Peter Steinberger, he was obviously loving it and and he did make that announcement. He's like, Open claw is now ready to run in your business. I mean, that's a little bit scary, isn't it? speaker-1: No. Absolutely. I mean, like, I think that the reason why open claw has been so interesting is that it is it feels similar to me like when we were working in ⁓ like GitHub Copilot in the IDE and we were asking questions and getting answers or doing autocomplete. And then suddenly you have this like agentic loop where it's ⁓ my ⁓ my goodness, it's going down a train of thought and going to do some work. I think when we're looking at the like assistant landscape, it it still feels very call response, you know? ⁓ in The claw just being able to kind of bring us that same, you know, chain of thought style, work with memory, et cetera, but across not just coding tasks is very, very exciting. And it helps whether it's knowledge workers, information workers, you know, folks that aren't day-to-day developers, I think understand the promise of what AI can actually do in like a consumer way. So now if we have a way to give you an experience that is fully personal to you. And have observability through like the Agent 365 stack, which you don't need to use our stuff or frameworks or agents, whatever you're ⁓ whatever agent you have you can monitor, and containment. I'm really hopeful that we'll be looking at more people using these like autopilots or autonomous agents or claws to experience the magic of ⁓ you know, that we all got to experience at like GPT three, you know, when chat GPT came out. Yeah. ⁓ because I don't think that that experience has been as broadly disseminated as the coding flows, where you ask something and it's ⁓ I called all the logs down from the server and I fixed it. Here you go. ⁓ we're all still very much on, I think, you know, go find me a reservation. It's like, you know, ⁓ this reservation service can't tell me if there's any times. It's like okay, great. You know, and then you kind of like go do it the manual way. ⁓ solving that gap and having all the sort of like compliance and governance that Microsoft's great at and our customers trust us for, will be, I think, a big boon ⁓ in the coming year. speaker-0: Well, I think that that's part of my my concern. Yeah. Right. ⁓ so isolation. That was one of the themes. ⁓ you know, you you mentioned that, your own containers. ⁓ so talk a bit about ⁓ about that in the importance of having that security model, a spectrum of security. ⁓ that came through, I think, pretty clearly a spectrum of isolation in terms of agents and the work we're gonna be doing. And particularly, yeah, what sh what should developers be thinking about and targeting? In terms of the portfolio that you've got. speaker-1: Yeah, yeah. So I mean, I think ⁓ like starting on the local side, these like ⁓ MXC, you know, is focused both on ⁓ like process level, network level, ⁓ some adaptations in the operating system to allow this on Windows. And then, you know, also bringing to, you know, Mac ⁓ and Linux the same underlying primitives ⁓ to give you that containerization locally. I think that for a lot of folks were I don't know, feeling the joy and friction of like always local agents. ⁓ or at least I do. Like I think it's a ton of benefit. But this morning I was asking my ⁓ asking my ⁓ agent to, you know, improve some ⁓ text that I was writing and I was like, ⁓ go run a rubric across this. And then I'm one of those people walking around the backstage with my laptop. ⁓ god, and I feel like a dummy. And it's just because we haven't been able to ⁓ like have that same experience in the cloud. speaker-0: Keep it open. speaker-1: I do think now with, you know, ⁓ the the host environments that Azure can provide for that, or like Azure hosted agents if you're actually gonna build a full agent and do that. I'm excited to see the containerization be ⁓ not the only reason why you're gonna go to the cloud, because in many ways we've all optimized for local because of the lack of containerization, the ability for these agents to feel magical and have full access. Now that we're bringing ⁓ containerization to local, I think that it will drive us to more seriously look, should this be local or should this be in the cloud? Because either way we have to containerize. Yeah. You know, it can't we can't just sort of take for granted this thing can run roughshod through everything that's on my computer anymore. ⁓ and in that way I'm hoping that I'll have containment, but I don't have to walk around with my laptop open ⁓ for coding tasks or for like my little small tasks. I can just you know, shoot a message or use remote control and could run on my laptop because that's the modality. speaker-0: And you know, you're gonna have like ⁓ speaker-1: I really want one of those. Absolutely, absolutely. speaker-0: Absolutely right. So yeah, that was that was ⁓ interesting news. ⁓ w let me let me ask you, like how much lead time ⁓ you know at this event did you I mean, was any of this news like when did you when did you when did you know about the My stuff? Like when did you know that speaker-1: Yeah. Well you Are you asking like how much of this news was produced for the event? speaker-0: Well it I mean there was a lot. It w there's like speaker-1: I mean like so I think the the like speaker-0: Was it was it like, ⁓ Kyle, just just one more thing or you know like ⁓ speaker-1: Like so ⁓ this is the first time when I've ⁓ you know been a part of build in this context versus just the GitHub context. And so, you know, I would say three-ish, four-ish months ago, we kind of went around to the company and said, like, well, I said, what do you got? You know, and like why is the developer gonna care about it? Like, how do we ⁓ focus as much as we can on like what's usable, but still be able to paint a picture of where we're going, you know, so it's not just about the tools we have today. and in that process, got to see some of these incredibly cool projects that ⁓ are both kind of pointing us towards the future. But I think the demo that Stevie did is like a perfect example of something that is actually useful, you know, ⁓ versus ⁓ some of the, you know, the era of either concocted or parlor trick, you know, AI demos where it's like, I want you to go book a plane ticket, you know? ⁓ and so n everything that you saw today, whether it be the MAI models or the GitHub Copilot app, Discovery, Foundry hosted agents, toolbox and foundry, which kind of got a little bit of a passing mention up front. That's all work that was ⁓ like well underway or like completed. You know, some of the stuff that we talked about ⁓ you know recently were fully delivered to general availability. The goal here was to kind of stitch it all together as best as we could to paint a picture of you do have the full stack, like you said at the beginning, ⁓ and We really want to invest in that, you know, kind of developer experience and possibility. Like I'm like was really excited about a lot of the ⁓ improvements that the Windows team brought through Windows for devs. I use like I've used my Mac. I've been a Mac user for a long time. I've gamed on PCs, I have a surface book, and I switch between the two because I'm responsible for all developers. You know, I have my homework box too, so I can like really cover all my bases. and you know. speaker-0: I use Windows now. speaker-1: Like jumping into Windows ⁓ after a Unix based operating system user for quite some time, running a like a Unix command Rutil and it not working was it's just like a piece of tax for me to switch over. And so the thing that makes me excited is, you know, the announcements today may not convince someone that's never like used a PC as a developer to just be like, ⁓ this is it. Now is the time. But if you do try one, I think there's the WSL stuff, there's PowerShell now. speaker-0: WSL stuff? Quick, quick, what was the WSL news? Like we can't expect everyone knows it. speaker-1: Sorry, yeah, sorry. So like yeah, I mean WSL too ⁓ has improvements now with like you can use the GPUs natively through it, you can have Kakuda support, has containers now so you can better containerize ⁓ your ⁓ your environments. It's got comfort shell, which allows you to use, you know, homebrew, Starship, ⁓ ZSH, like whatever is your normal setup ⁓ in those environments versus always feeling like ⁓ it always kind of felt like a little bit like a cloud VM to me, you know, where I'm just like on an Ubuntu box or whatever, you know, distribution I was using. And so really deep investments into that side, and then being able to use WSL from ⁓ across Windows. So if you know, instead of only being able to pull it up in a terminal, being able to call into it via your other apps, your other processes a bit more easily. ⁓ and then in Windows just kind of all up to being able to have your Git, you know, repo information in File Explorer, ⁓ being able to use, like I was mentioning, the utils in PowerShell because I'll still generally like open up, you know, a PowerShell environment and then, you know, try to do something and realize, ⁓ being able to have all that has been really, really good. and then like ⁓ you know, Kayla showed here, being able to reach into the GPU and use a local model is just cool. ⁓ you know, ⁓ and it works like it says on the tin, at least on like these, you know, newer ⁓ devices ⁓ in the in the Windows landscape. Yeah. speaker-0: So talk about devices then, 'cause I think that's one of the I to be it to be fair, I think is it it's it's slightly annoying. So we've got these hardware reveals. Yeah, I don't And then it's like and you won't get them speaker-1: Charlie McMarton, yeah. speaker-0: Yeah, that you're saying, Yeah, I'm on the waiting list as well. And I'm like, come on. So so w tell me about because I ⁓ that's certainly been something that I've been arguing that you you've got to have like these beast machines to make people be like, Okay, I just wanna I just wanna see what it's like from a p performance perspective. So ⁓ there were probably well, there were a couple of announcements there, but obviously the the NVIDIA partnership around Spark and what you've done there what what what's that about? speaker-1: Yep. ⁓ Yeah, I think, you know, similar to what you're saying, it's it it it's the ⁓ it's the confluence of, you know, hardware, software operating systems coming together that I think really is what makes this feel special. Obviously with N1X and you know the Service Book Ultra or the N one X dev box, you know, that we showed off. ⁓ That is about ensuring I have the hardware necessary to actually run these models effectively. Or like ⁓ there's a there was a cool demo when they were showing it to me ⁓ a while ago of you know, doing all of that and they were playing the ⁓ new one of the newer video game releases that were coming out on the same box without anyone seeing it, because it's all the same tech. ⁓ and I was like, this is great. I want to work device I can play video games on at night, but you know, like being able to have all that hardware, but then also really, really looking at the experience of using Windows eleven, you know, just all up with Proven and team. You they've been sharing a lot of the updates and improvements there. But I think I'm most interested in the developer experience side of that. speaker-0: thought I thought that was undebate actually today. But I'm gonna the the the the what you know what I'm like I'm in my mind I'm thinking the the developer distro of that I felt like ⁓ y the the the the question of calm and having something that's not getting in our face and you know it's like the the anti clippy. The the you know not not to have copilots in your face, just to really think about the experience. I thought that that you almost could have speaker-1: It's fine. Yeah, yeah. speaker-0: You I would have liked to have seen more of the articulation of calmness ⁓ and not getting in a developer's way. So where did that work stuff? speaker-1: Yeah, so like the calmness idea is ⁓ like represented here with the developer config, but realistically, like that is something for all Windows eleven users. Like they're improving Windows 11 just to be what you know, like they say, more calm. ⁓ I can turn things on if I want them, but otherwise not. the main thing with the developer config is a great example of all of our teams working together. You know, like the GitHub teams working and talking about the devices and Windows 11 and saying, okay. Well, when I'm starting a computer as a developer, this is what I'm gonna go do next. And making sure that that first run of jumping into a terminal is my first, you know, point of entry, I don't have to set up a bunch of config, or I don't have to go into like a totally like unopinionated ⁓ terminal flow or be you know, only able to use one agent versus a multitude. So there's this huge ⁓ foundation of the Windows 11 team ⁓ Just responding to all the user feedback, you know, in the last few months. And then they realize too that there's this giant cohort of developers ⁓ that have unique needs. And how can we build for those people in addition to every ⁓ every user? So I'm glad like Calm came through, but the real thing we were trying to land is yes, it's a config now, but you can imagine, you know, in the future that this is just, you know, whenever you whenever the next, you know, a Windows ⁓ update happens. You're developer and this is so this is the config we'll we'll have for you versus needing to go run it yourself. And this is where we talk about this at GitHub all the time, right? ⁓ there's only so much ⁓ prescription that I recommend to GitHub or Microsoft all up on what developers do want. And so, like, you know, the great developer config or my dot files may not match yours in the next person's. And so how can we find that happy medium of enabling devs, but not kind of creating an environment that every dev is gonna be ⁓ using and they go, well, this is not how I want my box set up. ⁓ that clean slate to work from is the goal and we'll see now with this config, you know, how closely we hit the mark. speaker-0: Okay, so talking about meeting a broader set of developers where they are, ⁓ there was actually some GitHub news as well. So we've got the new app. ⁓ one of the things that I've been really thinking about in my research and sort of looking at what's happening is there's an old saying they used to that it came out of Sun Microsystems actually, and that's that innovation happens elsewhere. And and today I feel it's like software development happens elsewhere. So software development used to happen in the developer's editor. Like that was the thing, right? ⁓ now it's happening all over the place. There's no reason, you know. I mean, if you look at the work that that, you know, an Atlassian or a linear software development could just happen in issues management. It can happen in any context. Software development's gonna be all around us, different sets of of users and everything else. Now you like I I th that was one thing that really interested me about the new what what is the new Copilot app? Yeah. And and is it part of this what we're actually yeah, software development isn't really happening in the editor anymore? speaker-1: Yeah, I mean, so the new ⁓ the new GitHub ⁓ copilot app is ⁓ like an agent first experience where ⁓ I'm gonna be ⁓ I'm able to like start projects or bring in my repositories, run a multitude of sessions where Git work trees are handled under the hood. I have a like a workflow infrastructure so I can run automations on my project or across my projects. I can respond to events that happen on GitHub and run automations on my local computer. And I can also use that same technology and use cloud posted agents. And so I can run and automate those like in the cloud. It provides me the ability to work across, you know, like on the weekends three or four different projects simultaneously and still take advantage of all of the sort of underlying, you know, plumbing and tech of GitHub. So create a PR and then I can say for most cases, like if I'm only working with two people, I'll say great, like agent merge, it'll go review, it'll respond to reviews. And ⁓ it'll resolve them. ⁓ or if I'm working with another person, like I'm currently trying to submit ⁓ functionality into the app itself. ⁓ and to be clear, I'm not like day to day coding anymore. ⁓ and so ⁓ like the thing that I'm doing is it's like okay, I'm using the full pipeline that they have. Like that was the thing that I found so compelling is because obviously the team that's building the app is using all the functionality. They have a bunch of review steps, a bunch of things. So I'm like not that guy because I do not I I'd built portions of GitHub's app a long time. I do not submit PRs to GitHub because I don't want to be that guy. But at least on these newer projects that are starting from that agent flow where it's gonna, you know, it creates an architecture document, does a review of that well before it gets to building my little feature. The app is encapsulating all of that. The thing that I found it's very new and we showed it off kind of quickly with Cassidy ⁓ did a quick demo of like opening a canvas. speaker-0: Great is Cassidy by the way. I mean you knew that before she came, but speaker-1: She's the best. She's the big ⁓ yeah, absolutely. I was so excited ⁓ that she joined us. But like she does a demo of, you know, opening a canvas, ⁓ and I'll explain that like, you know, doing like a thumbs up ⁓ and it, you know, merges the PR. The canvases are essentially a way for you to build ⁓ two-way communication via effectively like HTML, JavaScript, CSS inside of the app with your agent. Yep. And so I feel like ⁓ I've created, we've all created some apps that effectively are just taking something an agent does and driving back to the agent. ⁓ We use these canvases internally to do a lot of the sort of like prioritization and like early review or hmm, a bug came in. Like how can we break it all apart and explore that and not lose the context in just like a chat session where I'm kind of waiting for the agent to come back. I can actually say, hey, are there any other issues like this? And we'll just show me all the issues and I can attach them. I can be like, okay, this these two are clearly the same. This one's not. Having that experience where it's not just about being able to build a great UX for it, it's about when I'm using that UX, my agent knows and can start acting. And vice versa. My agent could say, ⁓ you know, have you considered this other ⁓ like interaction modality? You know, and add ⁓ I don't know, like say a slider or something on the age. ⁓ that's a really interesting ⁓ flow for the app that I'm excited for us to kind of see what everyone builds, because those are just code, you know, and code can just be shared. ⁓ but I do think that if you're the type of person that was using the CLI and either we're only working in one session, you know, or we're multiplexing like crazy and, you know, we're running a ton of sessions, ⁓ I just find the app to be so much easier to be able to do, you know, multiple different ideas ⁓ at once with the containerization we talked about, with the ability to do it in the cloud. ⁓ and then, you know, we showed the ability to both run Edit, ⁓ I think we talked about it. You can still just drop out to an IDE or deploy, you know, all from the single a single application. Yeah. speaker-0: Yeah. There's an awful lot to talk about. ⁓ I think ⁓ yeah, we could go down the whole get work trees ⁓ and store anyway. Worth watching the keynote. ⁓ I think I maybe to finish up a little bit. Right. One of the things that you know Sarcha's always been about, that Microsoft has always been about has has been a claim of of ⁓ realizing human potential. ⁓ this is an exciting time, ⁓ but it is also a really scary time for a lot of developers. ⁓ continued layoffs ⁓ across the industry, more and more waves of them. ⁓ you know, people that over last twenty years would always be able to get a new job immediately or finding that's not so easy. there is some definite fear, uncertainty, and doubt ⁓ in markets. So today I think Satra said some things about, you know, how models needed to support human potential, ⁓ how AI needed to support it. And that's right. What can you do? ⁓ and you know, obviously the GitHub touches pretty much every developer's life. Microsoft, there's a whole other set of services. How is the role not yeah, how do you see the role of the developer changing? And how do you make sure that you're supporting ⁓ I think a such a third rather than replacing developers? How does that how does it work from an organizing organizing principle? Yeah. And also then that how do you like How do market that and make it clear that you're on the side of the developer? speaker-1: Yeah. Yeah. I mean, you know, from the earliest time when, you know, we chose to, you know, say, okay, we're gonna call it GitHub copilot. That that wasn't the principle. The principle's always been about enabling developers to do more. I think the the piece that we've really stuck to is developers are in the loop of everything I've talked about today. Like there's every single step. We're letting you decide. You choose. If you want to do an agent merge, okay. On what pieces are you okay letting the agent go take control, or what pieces do you still want to retain control and we'll let the agent do the rest? That principle is core to the software flows that we build with at GitHub and Microsoft. I think the thing that we're we're like investing in and we need to do more of is how can we help developers be able to use these AI native and powered tools effectively and not need to keep up with the just volume of what's coming down the pipe, ⁓ because there's just so much. And I think that's a big part of ⁓ the the problem. When we're talking to developers about this, like we have this like concern that there's just so much to go learn. You know, in a speed in which I don't think we've ever really experienced like the last couple of big platform shifts have taken much, much, much more time to go from assembly to a higher level language and so and so forth. Now we're talking weeks to have a whole new idea come and leave. ⁓ and so our focus and I think our commitment is we can build that ⁓ we can build great tools and then we can make them easy for you to intuit about ⁓ and not have to know every single model when your employer is saying, Hey, choose the right model for you. Great. How can we help you do that? ⁓ how do we help you review your code with rubber duck, you know, to let you help have that kind of side by side comparison across models? Those small things I think add up to a big thing, which is still enabling developers to do their best work, not have to take the cognitive overhead that AI can sometimes bring us in the pursuit of doing more. ⁓ and then stick to our goal. We're developer first, we're about empowering them in the loop. ⁓ and then, you know, celebrating them when they're, you know, building and sharing whatever they whatever they ship with us. Mm-hmm. speaker-0: I am going to ask a last question actually. I thought I'd ask the last question. But I've got you here and ⁓ I I've got to ask you. So, you know, we're talking about all the new stuff. Yeah. ⁓ pretty clearly, you've got to stick to the knitting. Yeah. ⁓ as I've got you, like how are you going to deal with some of the challenges around reliability and and and some of the issues? Because best will in the world, w I I do feel I should I should be authentic. I should ask. Absolutely. So, you know, there is some market pressure here. What's how do you How do you deal with that from a North Star perspective? How are you gonna deal with ⁓ the the amount of load that agents are driving? How are you gonna fix some of these issues? speaker-1: Yeah. Yeah. I mean, so we've been working on this, you know, ⁓ really like deeply since around like last October. A lot of it is that we're, you know, increasing the number of CPUs that we need, as simple as that sounds. The scale is so quick, that's a huge piece. We've been rewriting and improving underlying core systems that do a lot of like the checking and policy where we have a step change moment and not just a very quick scale moment because suddenly, you know. The agents will all show up when rate limits go away and whiz bang boom across the industry, you know, we have a lot more work. So a lot of that has been happening underneath the hood. The core orchestration layer of GitHub Actions has been like very, very, very much improved with way more ability for us to launch ⁓ the various, you know, CPUs and boxes, box setups that you're looking for. Cause I know that's been a big ⁓ you know, ⁓ constraint for folks. As new model launches have come in, that's been another place. It's been hard, you know, because Each model has different characteristics. We have to land and at the full scale of policy. And so like we've improved that process a lot. ⁓ and we're, you know, making really good progress on getting to a you know cloud architecture versus just relying on our, you know, our own infrastructure. And so that includes Azure. That includes like we've always been had a multi-cloud and a physical infrastructure strategy. It's the demand of ironically, sometimes CPUs, ⁓ even more so than GPUs that have been ⁓ really, really big. So We know our like the first thing we're talking about every week in our meetings is about availability, reliability, how's it going, what are our customers saying, how can we ch meet with them? How can we talk to them? But realistically, we just need to prove it, you know, we need to do the work, we have to ship it. And then I think ⁓ I really want to get us back to the basics and I wanna talk about it. Like let's share the tech of what exactly we did, you know, ⁓ ⁓ so that way, you know, not as an excuse, but like as two developers do over beers or coffee, you know, explain the hard problem and ⁓ and what they did. So we'll have the first kind of sets of those posts going out from our leader, ⁓ technical leaders, you know, ⁓ speaker-0: People are hungry for that, so that's that's great to hear. ⁓ great. Well, ⁓ build, ⁓ twenty twenty six, quite developer-y, tons and tons of news. ⁓ it's been amazing to catch up with you, Carl. So thanks so much for joining us. Okay. Thank you all for joining us too. ⁓ if you got any comments, like, subscribe, all of that good stuff, and that's us. Peace out. speaker-1: Thanks for having me, Chase.